What is a prototype?
The Harvard Chan AI Glossary
In our classes, we often pause to explain the ideas behind AI and the language used to describe it. These posts grew out of those conversations. They’re for anyone who would like to understand AI better, with no technical background assumed.
A prototype is something created to examine an idea or answer a learning question. It might be a sketch, a clickable sequence of screens or working software. Its purpose is to make an assumption tangible enough to investigate. It does not need to contain every component of a finished product.
Suppose a team is considering an assistant that explains household bills. A prototype could show how a user uploads a bill, sees a proposed explanation and asks for clarification. The team might initially supply example answers manually. That can test whether the interaction is understandable without yet proving that an AI system can generate accurate explanations.
The distinction between real and simulated behavior should be explicit. A polished interface may conceal the fact that there is no connection to live records, no persistent storage or no model behind the response. Persistent storage means information remains available after the current interaction ends. These omissions may be entirely appropriate for a prototype, provided they do not mislead the people evaluating it.
A prototype should be shaped by a question. Are people able to understand the suggested explanation? Do they know which source supports it? Can they correct an error? Building every possible feature before asking those questions increases work without necessarily increasing learning.
Treat prototypes as learning instruments, including versions that can be discarded after feedback. Throwing away a prototype can be a good result when it has exposed an unsuitable interaction or clarified a better problem to solve. Its value need not lie in preserving its code.
A prototype differs from a minimum viable product, a deliberately limited product intended to test whether the value is real, and from production software that people rely on for continuing work. These labels can overlap in ordinary conversation, so describe what the artifact actually supports.
The public dental-care dashboard discussed in class was a good prototype question: could a visible sequence from attendance to treatment and follow-up help people examine where patients lose contact with care? Fictional records could make that interaction tangible without connecting to a real clinic. Such a prototype would not establish that the live records exist in the required form or that using the dashboard improves follow-up. It would make the next questions easier to identify and test.
If you would like to find out more about AI and improve your leadership skills, consider the AI in Health Care Certificate of Specialization from the Advanced Learning Academy at Harvard T.H. Chan School of Public Health. You can take an individual course or work toward the full certificate.
AI is a fast-moving field. This explanation reflects our class discussions and related material at the time of writing. Terminology, capabilities and tools may change.
For suggested additions or corrections, please email me at tpanch@hsph.harvard.edu.
Dr. Trishan Panch
Program Director, AI in Health Care
Harvard T.H. Chan School of Public Health — Advanced Learning Academy